Modeling Ex-Pump Fuel Prices in Ghana: Price Determinants and Comparative Evaluation of Statistical, Machine Learning and Deep Learning Models

Abstract

Ex-pump fuel prices are important economic indicators because their fluctuations affect transport costs, production expenses, inflation and household expenditure. In Ghana, accurate fuel-price forecasting is particularly relevant because domestic prices are influenced by refinery cost, crude-oil price and exchange-rate movements. This study examined the determinants of ex-pump fuel prices in Ghana and compared the forecasting performance of statistical, machine learning and deep learning models. A time-series forecasting approach was applied to biweekly data on gasoline, diesel and liquefied petroleum gas (LPG) prices, together with refinery cost, crude-oil price and exchange rate, covering September 2015 to June 2025. The first 80% of observations were used for model training, while the remaining 20% were reserved for testing. Sixteen forecasting models were evaluated using Mean Square Error and Mean Absolute Percentage Error. The findings indicate that refinery cost and exchange rate were the main drivers of domestic fuel prices. Refinery cost showed a very strong positive relationship with LPG prices (r = 0.98). Although gasoline, diesel and LPG exhibited broadly similar long-term upward trends, their short-term movements and adjustment patterns differed. Model performance also varied by fuel type. The Long Short-Term Memory model produced the best forecasts for gasoline, with a MAPE of 6.54%. The K-Nearest Neighbours model performed best for diesel, with a MAPE of 7.04%, while the Prophet model performed best for LPG, with a MAPE of 4.33%. The results suggest that fuel-price forecasting in Ghana should be product-specific and should incorporate relevant explanatory variables.